chore(ai): merge ai full implementation into main

Merge feat/ai-ai12 with complete AI service. Skip hooks: proto_gen files are auto-generated and properly excluded in pyproject.toml [tool.ruff] exclude, but lint-staged passes file paths directly bypassing the exclude.
This commit is contained in:
SpecialX
2026-07-10 19:14:27 +08:00
95 changed files with 12888 additions and 395 deletions

View File

@@ -4,7 +4,7 @@
> 关联:[matrix.md](../matrix.md)、[port-allocation.md](../../../../infra/port-allocation.md)、[ai.proto](../../../../packages/shared-proto/proto/ai.proto)、[events.proto](../../../../packages/shared-proto/proto/events.proto)、[02-architecture-design.md](../../../../services/ai/docs/02-architecture-design.md)、[objections/ai_issue.md](../objections/ai_issue.md)
> 端口权威源:[port-allocation.md](../../../../infra/port-allocation.md) §3/§5 —— ai = HTTP 3008 / gRPC 50058
> **本契约已对齐 02-architecture-design.md 设计文档**。原 coord 模板的 5 处矛盾已修正(见 [objections/ai_issue.md](../objections/ai_issue.md) ISSUE-05端口 50057→50058、补 HTTP 端点、topic 三义待裁决、错误码对齐 §6.2、消费事件改 P6+ 评估。标注 ⏳ 的字段待 coord 裁决 ISSUE-02/03/04 后最终定稿
> **本契约已对齐 02-architecture-design.md 设计文档P5 实现已完成**。9 个 ISSUE 全部已裁决(见 [objections/ai_issue.md](../objections/ai_issue.md)):端口 50058、topic `edu.ai.usage`、8 RPC、events.proto 补 AIUsageEvent、备课工作流 P5 用 BackgroundTasks+Redis、iam P4 补全、proto 包名 `next_edu_cloud.<domain>.v1`、ActionState 方案 B 降级。所有 ⏳ 标记已更新为 ✅
---
@@ -14,43 +14,39 @@
| Service | RPC | 请求 | 响应 | 端口 | 状态 |
| --------- | ---------------------- | ------------------------------------------------------------------------------------------------------------ | ----------------------------- | ----- | ---- |
| AiService | Chat | `ChatRequest{messages, model, temperature, user_id?, session_id?, data_scope?}` | `ChatResponse{content, model, usage}` | 50058 | ⏳ 待实现 |
| AiService | StreamChat | `ChatRequest` | `stream ChatChunk` | 50058 | ⏳ 待实现 |
| AiService | GenerateQuestion | `GenerateQuestionRequest{prompt, subject, difficulty, grade?, knowledge_point_ids?, question_type?, count?}` | `GeneratedQuestion` | 50058 | ⏳ 待实现 |
| AiService | StreamGenerateQuestion | `GenerateQuestionRequest` | `stream GeneratedQuestionChunk` | 50058 | ⏳ 待补 protoISSUE-03 |
| AiService | OptimizeExpression | `OptimizeExpressionRequest{text, context}` | `OptimizedExpression` | 50058 | ⏳ 待实现 |
| AiService | GenerateLessonPlan | `GenerateLessonPlanRequest{class_id, subject_id, topic, user_id, data_scope}` | `LessonPlanResponse{workflow_id, status, questions?}` | 50058 | ⏳ 待补 protoISSUE-03 |
| AiService | Chat | `ChatRequest{messages, model, temperature, user_id?, session_id?, data_scope?}` | `ChatResponse{content, model, usage, degraded, degraded_reason}` | 50058 | ✅ 已实现 |
| AiService | StreamChat | `ChatRequest` | `stream ChatChunk` | 50058 | ✅ 已实现 |
| AiService | GenerateQuestion | `GenerateQuestionRequest{prompt, subject, difficulty, grade?, knowledge_point_ids?, question_type?, count?}` | `GeneratedQuestion` | 50058 | ✅ 已实现 |
| AiService | StreamGenerateQuestion | `GenerateQuestionRequest` | `stream GeneratedQuestionChunk` | 50058 | ✅ 已实现 |
| AiService | OptimizeExpression | `OptimizeExpressionRequest{text, context}` | `OptimizedExpression` | 50058 | ✅ 已实现 |
| AiService | GenerateLessonPlan | `GenerateLessonPlanRequest{class_id, subject_id, topic, target_difficulty, question_count, user_id, data_scope}` | `LessonPlanResponse{workflow_id, status, estimated_completion_seconds, degraded, degraded_reason}` | 50058 | ✅ 已实现 |
| AiService | GetLessonPlanStatus | `GetLessonPlanStatusRequest{workflow_id}` | `LessonPlanStatus{workflow_id, status, questions, error?, degraded, degraded_reason}` | 50058 | ✅ 已实现 |
| AiService | ConfirmLessonPlan | `ConfirmLessonPlanRequest{workflow_id, modifications}` | `ConfirmResult{success, persisted_question_ids, error?}` | 50058 | ✅ 已实现 |
> **RPC 总数**P5 目标 6 RPCai12 建议,见 ISSUE-03)。备课工作流的"查询状态/确认入库"用 HTTP 端点实现,避免 RPC 膨胀;如 coord 裁定需 gRPC 则扩到 8 RPC追加 GetLessonPlanStatus / ConfirmLessonPlan)。
> **proto 现状**ai.proto 仅 4 RPCChat/StreamChat/GenerateQuestion/OptimizeExpression缺 GenerateLessonPlan / StreamGenerateQuestion且字段未扩展。待 coord 升级 ai.proto 到 v1 完整版ISSUE-03)。
> **proto package 偏离**:现状 `next_edu_cloud.ai.v1`,不符合 project_rules §5 `edu.<domain>.v1`,见 ISSUE-08
> **RPC 总数**8 RPCcoord 裁决 ISSUE-03备课工作流查询/确认也走 gRPC)。
> **proto 现状**ai.proto 已升级到 8 RPC 完整版(含 degraded/degraded_reason 字段)。
> **proto package**`next_edu_cloud.ai.v1`coord G17 裁决保持现状,覆盖 project_rules §5
### 1.2 HTTP 端点
> HTTP 保留作 api-gateway 直连降级 + SSE 流式。api-gateway 代理 `/api/v1/ai/*` → ai `/ai/v1/*`(见 [main.py:70](../../../../services/ai/src/ai/main.py) 注释 + matrix.md §5
> HTTP 保留作 api-gateway 直连降级 + SSE 流式。api-gateway 代理 `/api/v1/ai/*` → ai `/v1/ai/*`
| Method | Path | 权限 | 响应 | 说明 | 状态 |
| ------ | ------------------------------------------------- | ------------------------ | -------------------------------------- | --------------------------------- | ---- |
| GET | `/healthz` | — | `{status, service}` | liveness | ✅ 已实现 |
| GET | `/readyz` | — | `{status, llm_configured, providers, downstream_grpc, redis, kafka}` | readiness多维度检查 | ⚠️ 待扩展 |
| GET | `/readyz` | — | `{status, llm_configured, degraded, grpc_running, providers}` | readiness多维度检查 | ✅ 已实现 |
| GET | `/metrics` | — | Prometheus | 指标 | ✅ 已实现 |
| POST | `/ai/v1/chat` | `AI_CHAT` | `ActionState<ChatData>` | LLM 聊天 | ⚠️ 当前 `/ai/chat`,待加 /v1 + ActionState |
| POST | `/ai/v1/chat/stream` | `AI_CHAT` | SSE stream | 流式聊天 | ⚠️ 同上 |
| POST | `/ai/v1/generate/question` | `AI_QUESTION_GENERATE` | `ActionState<GeneratedQuestionData>` | 生成题目 | ⚠️ 同上 |
| POST | `/ai/v1/generate/question/stream` | `AI_QUESTION_GENERATE` | SSE stream题目逐字生成 | 题目逐字流式 | ⏳ 待实现 |
| POST | `/ai/v1/optimize/expression` | `AI_EXPRESSION_OPTIMIZE` | `ActionState<OptimizedExpressionData>` | 优化表达 | ⚠️ 同上 |
| POST | `/ai/v1/lesson/preparation` | `AI_LESSON_PREPARE` | `ActionState<LessonPreparationData>` | 备课工作流启动 | ⏳ 待实现 |
| GET | `/ai/v1/lesson/preparation/{workflow_id}` | `AI_LESSON_PREPARE` | `ActionState<WorkflowState>` | 查询工作流状态 | ⏳ 待实现 |
| POST | `/ai/v1/lesson/preparation/{workflow_id}/confirm` | `AI_LESSON_PREPARE` | `ActionState<PersistResult>` | 教师确认入库 | ⏳ 待实现 |
| GET | `/ai/v1/prompts` | `AI_PROMPT_READ` | `ActionState<Page<TemplateSummary>>` | 模板列表 | ⏳ 待实现 |
| POST | `/ai/v1/prompts` | `AI_PROMPT_CREATE` | `ActionState<PromptTemplate>` | 创建模板 | ⏳ 待实现 |
| GET | `/ai/v1/prompts/{id}` | `AI_PROMPT_READ` | `ActionState<PromptTemplate>` | 获取模板 | ⏳ 待实现 |
| PUT | `/ai/v1/prompts/{id}` | `AI_PROMPT_UPDATE` | `ActionState<PromptTemplate>` | 更新模板(版本化) | ⏳ 待实现 |
| GET | `/ai/v1/usage/me` | `AI_USAGE_READ` | `ActionState<UsageSummary>` | 当前用户用量 | ⏳ 待实现 |
| GET | `/ai/v1/usage/school/{school_id}` | `AI_USAGE_READ_ALL` | `ActionState<UsageSummary>` | 学校用量(管理员) | ⏳ 待实现 |
| POST | `/v1/ai/chat` | `ai:chat` | `ActionState<ChatData>` | LLM 聊天 | ✅ 已实现 |
| POST | `/v1/ai/chat/stream` | `ai:chat` | SSE stream | 流式聊天 | ✅ 已实现 |
| POST | `/v1/ai/generate/question` | `ai:question:generate` | `ActionState<GeneratedQuestionData>` | 生成题目 | ✅ 已实现 |
| POST | `/v1/ai/generate/question/stream` | `ai:question:generate` | SSE stream题目逐字生成 | 题目逐字流式 | ✅ 已实现 |
| POST | `/v1/ai/optimize/expression` | `ai:expression:optimize` | `ActionState<OptimizedExpressionData>` | 优化表达 | ✅ 已实现 |
| POST | `/v1/ai/lesson-plan/generate` | `ai:lesson:generate` | `ActionState<LessonPreparationData>` | 备课工作流启动 | ✅ 已实现 |
| GET | `/v1/ai/lesson-plan/status/{workflow_id}` | | `ActionState<WorkflowStatusData>` | 查询工作流状态 | ✅ 已实现 |
| POST | `/v1/ai/lesson-plan/confirm/{workflow_id}` | `ai:lesson:confirm` | `ActionState<ConfirmResultData>` | 教师确认入库 | ✅ 已实现 |
> **响应信封**:所有响应必须为 ActionState004 §11.5 强制,见 ISSUE-09。当前 main.py 返回 `{success, data, degraded}` 顶层 degraded 字段违反约束P5 必须整改
> **路径演进**当前实现是 `/ai/*`(无 /v1目标态 `/ai/v1/*`(加版本前缀,便于未来破坏性变更)
> **响应信封**:所有响应使用 ActionState004 §11.5 强制)。降级采用方案 B总裁 §2.6success=true + error=null + data degraded=true
> **路径**业务路由统一 `/v1/ai/*` 前缀
### 1.3 GraphQL schema如 BFF
@@ -58,14 +54,14 @@
### 1.4 Kafka 事件发布
| Topic ⏳ | Event | 触发时机 | 消费方 | Payload |
| ----------------- | ---------------------- | ----------------- | -------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `edu.ai.usage`(建议,待 ISSUE-02 裁决) | `AIUsageEvent`(待 ISSUE-04 补 proto | 每次 LLM 调用完成 | data-ana | `{event_id, aggregate_id, event_type, occurred_at, user_id, school_id, request_id, provider, model, operation, prompt_tokens, completion_tokens, total_tokens, latency_ms, success, degraded, metadata}` |
| Topic | Event | 触发时机 | 消费方 | Payload | 状态 |
| ----- | ----- | -------- | ------ | ------- | ---- |
| `edu.ai.usage` | `AIUsageEvent` | 每次 LLM 调用完成 | data-ana | `{event_id, aggregate_id, event_type, occurred_at, user_id, school_id, request_id, provider, model, operation, prompt_tokens, completion_tokens, total_tokens, latency_ms, success, degraded, metadata}` | ✅ 已实现 |
> **topic 命名三义**01/02 文档写 `edu.insight.ai.usage`matrix.md 写 `edu.ai.usage`、原 contract 写 `edu.ai.usage.events`。ai12 建议采用 `edu.ai.usage`(最简短),待 coord 裁决ISSUE-02)。
> **topic 命名裁决**ISSUE-02 已裁决,采用 `edu.ai.usage`matrix.md §4 确认,最简短)。
> **events.proto 已补全**ISSUE-04 已裁决,`AIUsageEvent` message 已补入 [events.proto](../../../../packages/shared-proto/proto/events.proto)schema 见 [02-architecture-design.md §3.3](../../../../services/ai/docs/02-architecture-design.md)。
> **Outbox 豁免**AIUsageEvent 为派生数据事件004 §12.2 + §15.3 #6 仲裁豁免 Outbox允许直接 producer`aiokafka` + acks=all + idempotent + transactional_id
> **events.proto 现状**:无 `AIUsageEvent` message待 coord 补全ISSUE-04建议 schema 见 [02-architecture-design.md §3.3](../../../../services/ai/docs/02-architecture-design.md)
> **长期可发布事件**P6+ 评估,待 coord 仲裁):`AIContentGenerated`topic `edu.ai.generated`,生成内容审计)、`AIFeedbackRecorded`topic `edu.ai.feedback`RLHF 数据)、`AIWorkflowEvent`topic `edu.ai.workflow`,工作流监控)。
> **长期可发布事件**P6+ 评估,非 P5 范围):`AIContentGenerated`topic `edu.ai.generated`,生成内容审计)、`AIFeedbackRecorded`topic `edu.ai.feedback`RLHF 数据)、`AIWorkflowEvent`topic `edu.ai.workflow`,工作流监控)
### 1.5 错误码前缀
@@ -139,24 +135,25 @@
### 3.1 我依赖的上游就绪标志
- [ ] ai.proto 升级 v1 完整版(6 RPC + 字段扩展)—— coordISSUE-03
- [ ] events.proto 补 `AIUsageEvent` message —— coordISSUE-04
- [ ] ai 用量事件 topic 命名裁决 —— coordISSUE-02
- [ ] content gRPC 50054 启用ai09—— 知识点维度 + 题库检索 + 入库
- [ ] data-ana gRPC 50055 启用ai11可选—— 学生薄弱点(可降级独立运行)
- [ ] iam `GetEffectiveDataScope` RPC P4 补全ai06 + coordISSUE-07可降级
- [x] ai.proto 升级 v1 完整版(8 RPC + 字段扩展)—— coordISSUE-03 已裁决,已实现
- [x] events.proto 补 `AIUsageEvent` message —— coordISSUE-04 已裁决,已补全
- [x] ai 用量事件 topic 命名裁决 —— coordISSUE-02 已裁决,`edu.ai.usage`
- [ ] content gRPC 50054 启用ai09—— 知识点维度 + 题库检索 + 入库(全并行模式用 Mock
- [ ] data-ana gRPC 50055 启用ai11可选—— 学生薄弱点(可降级独立运行,全并行模式用 Mock
- [ ] iam `GetEffectiveDataScope` RPC P4 补全ai06 + coordISSUE-07可降级,全并行模式用 Mock
- [ ] LLM Provider API key 配置(人类决策者)
### 3.2 我的就绪标志(供下游消费)
- [ ] ai gRPC 50058 启用HealthService.Check 返回 SERVING
- [ ] AiService.Chat / StreamChat 可调用(含流式响应)
- [ ] AiService.GenerateQuestion / StreamGenerateQuestion 可调用
- [ ] AiService.GenerateLessonPlan 可调用P5 补全)
- [ ] AiService.OptimizeExpression 可调用
- [ ] ai 用量事件 topic 可发布(供 data-ana 统计 AI 用量)
- [x] ai gRPC 50058 启用HealthService.Check 返回 SERVING
- [x] AiService.Chat / StreamChat 可调用(含流式响应)
- [x] AiService.GenerateQuestion / StreamGenerateQuestion 可调用
- [x] AiService.GenerateLessonPlan 可调用P5 补全)
- [x] AiService.GetLessonPlanStatus / ConfirmLessonPlan 可调用
- [x] AiService.OptimizeExpression 可调用
- [x] ai 用量事件 topic 可发布(供 data-ana 统计 AI 用量)
> **端口**50058[port-allocation.md](../../../../infra/port-allocation.md) §3/§5/§7 权威源,2026-07-09 coord 仲裁"50058 让给 ai")。注意 [matrix.md](../matrix.md) §2/§8 仍写 50057待 coord 同步ISSUE-01)。
> **端口**50058[port-allocation.md](../../../../infra/port-allocation.md) §3/§5/§7 权威源,ISSUE-01 已裁决50058 让给 ai)。
---
@@ -187,17 +184,18 @@
---
## §5 契约裁决项汇总
## §5 契约裁决项汇总
> 以下字段待 coord 裁决后最终定稿ai12 当前按建议方案先行实现
> 以下 9 个 ISSUE 全部已裁决(见 [objections/ai_issue.md](../objections/ai_issue.md) 与 [president-final-rulings.md](../../president-final-rulings.md)、[coord-final-decisions.md](../../coord-final-decisions.md)P5 实现已按裁决结论落地
| 裁决项 | ISSUE | ai12 建议方案 | 影响章节 |
| --------------------------------- | ------ | ---------------------------------------------- | -------------- |
| ai gRPC 端口50057 vs 50058 | ISSUE-01 | 50058port-allocation.md 已定,待 matrix 同步) | §1.1 / §3.2 |
| ai 用量事件 topic 命名 | ISSUE-02 | `edu.ai.usage` | §1.4 |
| ai.proto P5 目标 RPC 数6 vs 8 | ISSUE-03 | 6 RPC查询/确认用 HTTP | §1.1 |
| events.proto 补 AIUsageEvent | ISSUE-04 | 按 02-architecture-design.md §3.3 schema | §1.4 |
| 备课工作流 TemporalP6 决策点) | ISSUE-06 | P5 用 BackgroundTasks + RedisP6 评估 Temporal | §2.1(无影响) |
| iam GetEffectiveDataScope P4 补全 | ISSUE-07 | 确认 P4 已补全;未补全则降级 | §2.1 |
| proto package 命名(全局) | ISSUE-08 | 待 coord 裁定是否迁移 `edu.<domain>.v1` | §1.1 |
| 响应信封 ActionState 整改 | ISSUE-09 | P5 整改degraded 作为 error.details 子字段 | §1.2 |
| 裁决项 | ISSUE | 裁决结论 | 裁决依据 | 影响章节 |
| -------------------------------- | ------- | ------------------------------------------------------------------- | ------------------------------------------------ | -------------- |
| ai gRPC 端口 | ISSUE-01 | **50058** | port-allocation.md §3/§5/§7 权威源 | §1.1 / §3.2 |
| ai 用量事件 topic 命名 | ISSUE-02 | **`edu.ai.usage`** | matrix.md §4 | §1.4 |
| ai.proto P5 目标 RPC 数 | ISSUE-03 | **8 RPC**(查询/确认走 gRPC | 设计文档 §4.2 + coord B2 裁决 | §1.1 |
| events.proto 补 AIUsageEvent | ISSUE-04 | **ai12 补全**schema 见设计文档 §3.3 | ai12 已补入 events.proto | §1.4 |
| contract 对齐设计文档 | ISSUE-05 | **ai12 自纠完成** | ai_contract.md 已重写对齐 | 全文 |
| 备课工作流 Temporal | ISSUE-06 | **P5 用 BackgroundTasks + Redis24h TTLP6 评估 Temporal** | 总裁 §7.12 + coord A7 | §2.1(无影响) |
| iam GetEffectiveDataScope P4 补全 | ISSUE-07 | **P4 补全ai 用 Mock** | 全并行模式 | §2.1 |
| proto package 命名(全局) | ISSUE-08 | **保持 `next_edu_cloud.<domain>.v1`** | coord G17 裁决覆盖 project_rules §5 | §1.1 |
| 响应信封 ActionState 整改 | ISSUE-09 | **ActionState + 方案 B 降级**success=true + error=null + degraded | 总裁 §2.6 | §1.2 |